Jump to content


  • Content Count

  • Joined

  • Last visited

  • Medals

Everything posted by mmm

  1. My, "Strong MP focus", "Monetization", "No DLC", I don't like the sound of them all in one place... There's even no mention of custom SP scenario it got me worried. I won't complain about braindead AI without adequate command feature anymore but BI I beg you spare the bots. And it seems ARMA 4 is stuck in development hell. Don't tell me they can't get savegame working is the reason why they don't have any SP...
  2. mmm

    The Fictional MX Rifle series, why?

    Given UK outspend Japan and more than twice of Israel, guess they'll have to make do with even smaller rocks and shorter sticks and sail on rafts...
  3. mmm

    The Fictional MX Rifle series, why?

    Well there's no jumping at neckbreaking acceleration, only silicon and software with some plastic so... Uncooled thermal sight with bluetooth and embedded chips, soldier smartphone with GPU for inference workload, Microsoft Hololens Army Edition, with another 14 years to go that doesn't sound far fetch to me at all. With how much commercial tech there is the cost will be literally dirt, especially compared to personnel cost of first world armies. The accelerated decision making at ARMA speed will be the next big leg up IMHO. Not that I think it can ever be represented 1:1 for (expect)real life, probably won't serve the most popular game modes and broader playerbase well, whatever they are today. But say that's the reality in 2035, an ARMA 6 wouldn't be "realistic" "contemporary" milsim without those elements would it? It would at most be a historic sim of war on terror era. How much ARMA is a "milsim" or how much they want it to be one is totally up for debate OFC. But then again I'm betting even in 2035 there will be sim enthusiast enjoy the most accurate representation, even if that reality seen from today doesn't look like fun gameplay. Who know maybe VR will be the prerequisite. I was talking more about the preference of the general market. Personally I'd love to see a meticulously studied 2035 ground combat sim with every era defining technology represented(in meaningful fashion), whether that still play like an industrial standard shooter is of no concern to me(god help my pleb reflex peasant frame rate). For all I know it could be Combat Mission with boots on the ground perspective and a ton more detail. Coming to think of it I actually like the idea. Now ARMA 3 pushed it too far no arguement, there's no way to study 2035 with any reasonable accuracy from 2013. But that time frame for a hypothetical 2023 ARMA 4 still sounds okay to me, a reassessed version based on updated knowledge. I think lack of visibility of more recent changes in trends and the implications in land warfare unlike the ~2000s does make a proper study more difficult.
  4. mmm

    The Fictional MX Rifle series, why?

    Well that "2035" setting have always been more superficial visual facelift than gameplay changing representation of expected technological advancement, but if they(CAN in the first place that is...) do there's not much game to play... Gun pulls the trigger for you, only when the ballistic computer feels you're pointing the right way, and vehicle won't even have that twitch reflex test, it just require your signature to kill so to speak. Probably even footsoldier will carry a GPU that runs detection algorithm on the imagery collected by HMD, with things like this going on... Your target tag and red circle highlight? It's not a difficulty setting, it's an equipment function. And all that shared on 1 unified network instead of today's several layers of platform specific non-interoperable ones? The AI hive mind is a feature, not a bug then. You don't even need jet or artillery for abrupt death, loitering munition can do just that. If you've been observed and targeted you'll die before you know.
  5. I'm under the impression that the algorithm can be made to run on GPU, and inference is not that heavy. Obviously that kind of problem is for BI to tackle, assuming they have any interest in doing so. Exactly because how badly it runs as is, even this many years later, I hope they look into approaches that may cost them more upfront to implement.
  6. Well there won't be ARMA 4 anytime soon... I think you'll have time to flesh out your satellite feed to complete the package before it comes out at least. I wonder if the deep machine neural network learning expert AI system thing will help ARMA, with regard to navigation behavior.
  7. mmm

    We need Arma 3 on consoles

    I don't know of a precedence so far for mandatory mouse keyboard on console. There's yet to be a VR only major title other than half life, and sony can't peddle first party mouse keyboard can they? As I understand M+KB is still something of a controversy on console for its competitive advantage in shooter. While I don't personally see ARMA as looking for maximum competitive advantage in competitive multiplayer mode, there's no guarantee it won't be the popular mentality for console demographic. And to be honest I'll be a bit concerned if that steer ARMA in different direction. Reminds me of the "Argo" thing also... But I'll give you a better counter argument, "you guys don't have a phone?" "I mean a mouse?" I understand the barrier is very low. Thanks for reminding me of that fact. Actually anyone with hands on experience on that can probably offer some insight into what a potential future port look like.
  8. mmm

    We need Arma 3 on consoles

    What's running ARMA with JUST controller like? I mean you are releasing on console, it has to offer full functionality for controller only, you can't make M+KB mandatory. I'm not qualified to comment at all with single digit hours of controller experience under my belt and 0 console owned in my life time, but is it at all possible to do it without stripping features from baseline ARMA 3 standard? Say you do that with button combinations and multilevel menus how's ergonomics for controller then? And what's the "ARMA experience" console players looking for? Full disclosure I'm not familiar with how most other play ARMA. I jumped into ARMA 3 for the kind of experience in the earlier pre-release advertising material on the single player campaign, as seen here and here, and having been looking for it ever since(and learned how impossible it is in the process with the ARMA 3 we currently have on hand...), although I do have the self awareness that I'm in a very small minority. From the little I know though it seems diverse and very mod dependent. Say it is technically feasible, is it the exact perfect replication of the entire ecosystem but just running on standardized hardware they would like? In another word as if they're just like all running on PC but with identical spec? Or it is not the case but they already find the vanilla offering compelling enough, or with a few officially sanctioned mods?
  9. mmm

    you need supercomputer to play this game

    Let me check the recommended spec on store page really quick... Hmm, I5 4460 or FX4300. Do people from future have anything to say about this? Wot?! Platoon sized scenario on Tanoa? If you've played ARMA at all you wouldn't have made that kind of mistake.
  10. Well vanilla AI can watch direction, they don't know to put a waist height object between it and that direction or a position. I mean can I make your AI always take cover against a specific direction without knowing the existence of an enemy. You know there's contextual information AI can't use like mission briefing or terrain features suggest potential enemy.
  11. mmm

    Old Man: game mechanics

    1. Well it's pretty much "teleport" with shifting time of day. It's not like there are persistent entities simulated in the background at 1000x speed in lower fidelity during the skip. But on your question I don't see a way to teleport your vehicle as well, design wise it would have made roadblocks pointless if they did though. 2. As far as I know roadblock repopulate based on the distance you've moved away from. At a distance long enough for the objects to despawn, and move back within the spawn distance of checkpoint guarantees a regenerated checkpoint. It's no more than 6,700m at max. However every other sites are pretty much a one time only business, every kill is permanent. Checkpoints and enemy reinforcements spawns are the 2 forms of indefinitely regenerating kind of enemy, at some point you'll have little incentives to fight them anymore, as they only expend you resource. 3. Enemy reinforcement waves a created, base on what I see, upon the complete destruction of the previous wave. You don't run into a stronger wave(eventually escalate into CAS fighter) if you manage to elude the previous one or only partially destroyed. It's the only condition from my observation. As you can probably tell the level of "dynamism" is not that deep in this mission.
  12. Okay, good to hear the cover taking's intact. I'm not 100% sure what exactly you meant, I guess you mean your equivalent of the vanilla AI's awkward peeking/pie slicing around scripted corners in "danger" combat mode, regardless of presence of detected threat. But you did mention "under fire", I wonder if there's going to be a mean for player to insert the expected threat direction, in absence of already detected threat, say take cover against north west while moving cover to cover, or maybe a exact grid position on map. The "cover and fire position analysis" sounds like exactly what ARMA AI need. I know of nothing technical but that cover system also look like something dynamic instead of pre-flagged on objects. Wonder if that's aided by recent advancements of deep learning? That kind of tech is probably expensive and do nothing for the profitability of more traditional games, but for ARMA I think that's the kind of investment necessary to really justify the next entry in the series.
  13. That sounds very pleb friendly, thanks. That's unfortunate to hear, I considered your AI cover taking a core feature. In my ARMA fantasy land I think it could be dealt with when combined with a command overhaul, where the ai improvement provide the "corner candidate" positions, and let human player filter out the false positives by manually designate the position and facing in the 3D scene.
  14. Well there's probably not that much AI improvement to ARMA franchise over the last 10 years or so, not when compared to what you're about to do. The command radio menu hardly changed at all except for minimum integration of some new ARMA 3 features... Almost as with many things ARMA they're just destined to remain WIP for, well forever. I say let us see what the 80% solution looks like, what's really bothering the end user and what's not that big a deal even if it not perfect. Wonder if you'd recommend any number of particular scenarios as your AI's intended ideal usage once it come out? I'll respectfully push back against the idea, ARMA terrain is drastically more complex and not with the level of "AI friendliness" of average AAA title in mind, events occur at greater distances, and AI don't cheat the mechanics to nearly the same degree, broadly speaking every entity including you as player operate under the same rule. How AI attempt to gain the line of sight on the target alone may be more detrimental that what you're imagining. At present I say just use unit switch to check line of sight and use the default assign target function. I won't say "stealth" is not a thing in ARMA, but it's a lot more nebulous without the level of consistency that you can play with, and generally non existent after firing the first shot.
  15. Vehicles as I understand has little organic ability to initially detect dismount units, probably to represent the general "blindness". But that also means they won't be able to engage dismount units unless: A. They come under fire themselves which means they're automatically given enough information about attacker to retaliate. But then again when they do come under fire it's general from weapons that can kill in a single hit. B. They're in a group with units(generally speaking boots on the ground) that can actually detect dismount, in that case they become in my word "brain datalinked" with perfect information of the entire group. What one sees all knows. It's rarely the case though. Once an entity is know though I think AI has ability far beyond human to maintain track, I don't claim to know how any of this works being just another guy that plays the game...
  16. In all likelihood you're at 2/2 level which is the max level, there isn't another one. Can't say it's necessarily better for you, you'll get insurgents with 1 magazine in the weapon only and no spare because someone forgot to change the entire loadout instead of just the weapon.
  17. That sounds troubling... But then ARMA is the kind of on and off thing, I don't claim to be playing on weekly basis through out the years, I hope it is that kind. Is that going to affect(read totally destroy) my ongoing mission save? I know it's probably easier to just try load a save to find out... BTW I can't remember for sure, is it your mod that enables the command view(the long distance "satellite" 3rd person view) regardless of existence subordinate? If so it is not working anymore.
  18. This and the previous ARGO thing, I wonder what they expected from the side projects. They weren't sold for money, arguable not in states that can be sold for money either, didn't further iterate into paid product, and plugs were pull in short order. Call me pessimistic but I can only assume they found out both were beyond fixing early on and because how they're released the cost to terminate is low. And nobody gets to see what the follow up to the experiments look like. While doing major overhaul on singleplayer defies logic in most games, I'd say dynamic full map SP in ARMA 3 is an exception. Those mods are generally results of years of iteration and fixes.
  19. Guess with the latest announcement this will be the final state this mission will remain in? Wonder how's the stability these days. With 2 sessions of dozens hour in then rendered impossible to proceed due to technical problems I'm really deterred from trying again. Maybe I should just stop hoarding and sight seeing and speed run this just to see the ending whatever it might be. When they say they've been "taught a lot of good lessons", I sincerely hope they meant it literally, not as a way of saying "we'll never try that again". This official "open world" campaign was something that I had eagerly anticipated with high hope but tempered expectation since first announced(man that's back mid 2018, with that gestation period can't imagine it was smooth sailing). Personally I would have hoped it to be something more and of a bit different interpretation of "open world" than what it end up being, but on the other hand it is more or less what the rational part of me expected under the circumstance(although I had expected something a bit more performance friendly and less prone to game breaking bugs). Hopefully in the future they'll draw inspirations from the community dynamic campaigns like DUWS, Antistasi or WLA:Amalgamation and others than I didn't have the chance to find out about. And take another shot at the concept when circumstances are more favorable.
  20. It just came to my mind now somehow, but wherever you're ready to have an updated showcase of your superAI demo video, wonder if you can include a segment demonstrate an AI squad doing something like THIS? Say like they pretend they took unexpected contact then just assume positions that human would, that's really going to leave an impression if you could. Also I wonder how much control over the position they take I have as a player squad leader.
  21. Very astute observation, those would be aptly classified as "tank destroyer", actually I don't know what else those are capable of other than fighting tanks(in an asymmetrical manner). As I understand "tank destroyer" historically carried a tank gun only for as long as recoilless shaped charge weaponry wasn't mature. Since then it was possible to cut down the cost of chassis significantly. Wheeled (modern)assault guns aren't generally meant to fire many sabots so they better be able to reflect their intended real life role. I went on to explain how I believe HE lethality modelling is too lacking to do so and how similarly inappropriately modeled coax/RCWS overshadowed high caliber HE when in real life there's not much overlap between the 2. 2x30mm autocannons, 4x SALOCS ATGM in the same weight class of hellfire(in contrast of more common ground launched ones) that's on the other hand almost impossible to reload in combat due to launcher configuration, unlike say TUA TOW, Bradley or their own 9P149, 2 more independently targetable bow GMG, armored similarly to MBT with less protected unmanned turret whose destruction doesn't translate to a catastrophic kill, and no troop compartment. That's the most distinct addition I could think of without major platform update to ARMA 3. I'd argue there's more distinction than between the MBTs if viewed under the same light. Of course there's the problem with the limited vehicle simulation fidelity it may not be able to reflect all attributes of such vehicle, and the delicate faction "balance", somethings that's a more serious consideration these days I guess.
  22. mmm

    The effect of weapon on AI accuracy

    My impression is that dispersion would not have been THAT significant if not for the AI dispersion coefficient that amplify if for a few times for AI usage, so it becomes likely the main source for weapon hit probability disparity. Just for example your previous test here seems corroborate the theory well. Those with strong performance in hit probability department are all configured with superior dispersion. While for real human player a flat trajectory and short flight time brought by higher muzzle velocity is likely more important than a moderate dispersion difference, as exemplified by the SPAR-16 Aka HK416 10.4" which totally defies the expectation and grossly outperforms a CTAR with otherwise mostly comparable external ballistics. Edit: As for the discrepancies between different AKMs, just off the top of my head, it might also be helpful to look at AI dispersion coefficient. I don't have access to GM but I assume it might be configured somewhat different from established standard there. RHS obviously adopted automatic fire for AI usage, which adversely affects killing efficiency as far as current ARMA AI firing simulation goes. I'm certain only the first round within a burst has any chance of hitting the intended target.
  23. I don't see many problems with vanilla AI accuracy in general or at least I don't have in mind a better place than it is today, it's obviously circumstantial depending on weaponconfig, units skill, difficulty aim skill etc, but I feel they can easily match human accuracy and in some ways far exceed by a wide margin, like say against moving targets, while being reasonably scalable for different "needs". There were some very informative tests done by @nkenny on AI accuracy, despite title it actually informs a lot on AI accuracy in general, especially with the most common assault rifle weapon class. Doesn't mean i want AI to get significantly less accurate overall as exemplified here, which I believe needs addressing as well(but probably won't be). The real problem I see is how AI handles re-detection after having initially spotted a target once, and their detection at closer distance regardless of other factors that affects human spotting, especially combined with their close range precision at way beyond human level. Also as far as I can tell, after a few quick turns and sprint, it appears player take some half a minute to recover the from extra weapon sway even with some of the most lightweight weapons, supposedly due to the weapon inertia. To this day I'm not sure if it is intended to be THAT long, but if you don't have the terrain to go prone on or deploy your bipod to quickly eliminate the sway, that's the time until you can deliver effective fire beyond point blank after coming out of movement, which is a deadly amount. AI don't appear to have that, or at least as bad as human players'. Also by: Can you elaborate a bit? do you mean you externalized the accuracy settings fully without the vanilla accuracy related setting having any effect? What happens to weapons configuration for AI, as you may know not all weapons are created equally for AI?
  24. I still don't get... Doesn't any kind shooting involves pointing at a specific azimuth/elevation, what's happening with your AI at 8:52 in your video, that's AI shooting right? Clearly with the strafing ability and kind of firing discipline and accuracy not present in vanilla? The loss of idle animation that has no real function doesn't bother me a bit, I don't even consider it any sacrifice for the kind of result you produced. As for stance adjustment, right lean is more or less functionally interchangeable with right adjust when used independently. Left lean on the other hand is fairly pointless without combining with a left adjust, which is in actuality a shoulder swap. You substituting left lean with left adjust can really improve AI cover taking even used on its own, especially on those walls with slanted cut on the left side.
  25. Hope your "satellite feed" revamp is also moving along well, as the vanilla command will definitely be insufficient for your AI. Hmm, do you mean when AI's firing it's using vanilla AI controlling the unit? But in your Demo the AI was strafing while shooting, something they never do? Is it correct for me to interpret that this is the kind of problem that could be handled a lot more easily say if you were a BI dev, but now you have to resort to workarounds and gimmicks?